[SunRescue] Anyone need some computing time or shell?

[Joshua D. Boyd]

On Fri, 16 Mar 2001, Reagen Ward wrote:

> RMS is a strange duck.  He and I exchanged email back in 91/92, but once
> I suggested that I write a front end to emacs to make it work like vi,
> he quit responding.  No sense of humor.  Actually, I'd say that he lacks
> any sense of humor whatsoever, not just when it comes to emacs.

Hehe.  I bet someone already wrote such a front end.

> > Anyway, I am always annoyed at how hard it can be to get free software
> > working on non GNU/Linux machines.  Most projects are tightly tied to GCC,
> > which is OK when trying to port to Solaris, but not OK when I'm trying to
> > port to Irix where the GCC port is clearly inferior performance wise to
> > the MIPSpro compiler.  I've sent a few diffs away cleaning things up to
> > pass by MIPS compiler, but they have always been ignored with no
> > explaination.  I'm getting real sick of this, especially since I intend to
> > buy an SGI O2 this year, and would like to make it my primary machine.
> 
> That compiler gives the best warning info, if you ask me.  I love it.
> It really me helps clean up code.  My only SGI is an Indigo2 with an
> R4400 (ZX gfx if anyone cares), so it's hardly a screamer, but it's got
> good I/O and a really snappy display, so I regularly use it as a
> desktop.  I do try to pump almost any C I have through the compiler on
> it just for the warnings.

I'm sure that the SGI compiler is more correct.  It is still a real pain
in the but when trying to compile cross platform code.  

Previous to this year, I would have gone for an Indigo2 with Max Impact
and an R10k chip instead of an O2.  For polygon performance, FPU
performance, etc, the Indigo2 would be both faster and cheaper than the O2
I plan to get.  However, older O2s come with video i/o hardware, and it is
a cheap add on for new O2s. The real reason though is that the O2 blows
the Indigo2 away in texture performance.  My current main projects have
more to do with 2d or 2 1/2d graphics with motion video, and I figure the
O2 is the best bang for my buck on those tasks.  Before I just wanted an
Indigo2 because SGIs were cool and I was all about 3D.  I'm still all
about 3D, but I think I have an idea to make money in 2D now.  I think the
2D effects software market will be easier to break into than 3D animation.
I want to make an affordable video effects package for the O2, then try
and get HP to support me in doing a linux version (they have the best 3D
and 2D performance under linux, but they get it buy running the HP-UX
Xserver on linux and using their proprietary cards), then maybe an NT
version.  By my calculations and tests, the type of NT video cards needed
cost more than the O2 will, but the price on those cards are falling fast.


> > On a slightly different note, I'm sick of all the OSS projects (like
> > Panora and GMAN) that talk about how they are going to be the best
> > renderer yet, and yet then never get past the simple ray-tracing stage.
> > If all I wanted was a good raytracer, I'd get better results out of
> > POVRay.  And person with the simplest programming skills should be able to
> > write a raytracer if they wanted too.
> 
> How well does BMRT work?  I used to play with Renderman.

BMRT is a bit slow.  Other than that, it lags a bit behind prman, but if
you can't afford prman ($5k per CPU) then this isn't a big deal.
Currently, the main short comings are that BMRT reportedly can't use more
than 2 gigs of memory, and is isn't quite as memory efficient as prman
(prman on solaris has the same 2 gig limitation, but Pixar limits all
the process on their render farm to a gig of RAM reportedly anyway).
Standards wise, subdivision surfaces are buggy under BMRT, and last I
hear, RiPoint was supported (the RiPoint structure is a more memory
efficient way to do particle type things).

For someone on a budget, BMRT is just the thing.  At today's prices, if
BMRT is fast enough, just enlarge your renderfarm.
 
> My main complaint like that is with the so-called CAD and modelling
> packages.  They almost all suck, with the exception of Cycas (non-free,
> but cheap) for CAD and AC3D (non-free, but cheaper) for simple modelling.

I've given up hope on those for the most part.  The problem is that these
packages are being written by programmers who don't understand how the
artist should work.  I've spent some time working on a simple modeller and
a simple animation tool that I can testify that getting things to work
smoothly is hard.  I'd mainly focus on animation since there are starting
to be a few modellers that handle one type of model handily (spatch for
curved surfaces, almost any quake2 character modeller can be used to make
decent subdivision surfaces if you can live without previews, and
there are a few blob modellers that are usable), so I plan to mainly focus
on animation once I get of this 2D kick.
 
> I spent about 5 months trying to find a 3D CAD package that worked well
> under BSD or Linux, preferably under Solaris, and didn't cost over
> $1500.  I found exactly one, and my wife is still waiting for the
> manuals to arrive.

I assume you are talking about blender?  Blender is a far cry better than
all the free packages I've seen, and it is better than most of the cheap
packages I've seen (I hear that a lot of new packages are out that are
better) from usability. Still, I don't think it holds a candle to
Lightwave and Softimage. 


--
Joshua Boyd